CN103324921A - Mobile identification method based on inner finger creases and mobile identification equipment thereof - Google Patents

Abstract

The invention provides an unrestricted mobile recognition method based on the inner finger stripes, which uses the camera of the mobile device as a collection device to collect hand images, transmits the collected samples to the server through the network, and automatically detects the hand area , inner finger stripe area location, inner finger stripe feature extraction and feature comparison with inner finger stripe samples in the database, so as to realize identity recognition based on inner finger stripe biometrics. The present invention also provides a mobile identification device for implementing the mobile identification method based on the horizontal stripes of the inner finger, including: a collection module, a transmission module, a preprocessing module, a processing module and a decision module. It is innovatively applied in the mobile environment and can adapt to changes in different backgrounds, lighting, postures and viewpoints. It has a certain tolerance for misalignment and a high matching success rate. It is a convenient, efficient and reliable identification system. It has good application prospects in the security field and other advantages.

Description

A mobile identification method and mobile identification device based on inner finger stripes

technical field

The invention relates to a mobile identification technology, in particular to a mobile identification method based on inner finger stripes and a mobile identification device thereof.

Background technique

With the popularity of mobile devices such as tablets and smart phones, we are dealing with confidential official affairs and personal privacy affairs on these devices that require strict identity authentication. Work is also becoming more and more frequent. However, the traditional method of identifying an identity by setting a user name and a password is not unique and is easy to be forgotten, cracked or stolen. Unlike PIN (Personal Identification Password), biometrics will not be forgotten, lost or stolen, and cannot be easily copied or shared. Therefore, biometric-based identification systems have great advantages over traditional password identification systems.

Now there are many widely used biometric identification systems, such as fingerprint, iris and so on. The existing biometric identification technology has the following points:

First, most biometric technologies rely on professional collection equipment to create a controllable sample collection environment to simplify the recognition algorithm and improve recognition accuracy. Even if some ordinary cameras are used, there are still certain requirements for the environment, and the recognition accuracy is not high. Therefore, in an uncontrollable mobile environment, it is difficult for existing technologies to obtain good recognition results.

Second, recognition methods based on behavioral features, such as gait, voice, signature and keystroke typing, etc., have low recognition accuracy and are easy to be imitated. Therefore, in view of the hardware conditions of existing mobile devices, it is generally considered to use a camera to collect image information such as palmprints and inner finger prints as identification basis.

In Human and machine recognition of faces: a survey, a face-based mobile recognition system is proposed, however, the system requires a large number of different lighting and Huge database of captured samples under pose conditions. For palmprint recognition systems, they often require very high-resolution pictures to obtain sufficient structural information for identification. However, when the posture changes, the high-resolution will cause the pictures to be significantly affected by the projection transformation.

It is mentioned in Illumination ratio image: synthesizing and recognition with varying illuminations (illumination quotient image: synthesis and recognition under variable illumination), compared with voice, face and palmprint, fingerprint stripes are more suitable for movement due to its natural characteristics biometric system. In addition to the smooth surface, the horizontal fingerprint also has rich structural information that is difficult to imitate, and because it is in a small area, its projection transformation is small or even negligible. That is to say, under the condition of low resolution, using fingerprints for recognition can get higher accuracy than using palmprints.

Among the currently proposed biometric identification methods using fingerprints, almost none of them have solved the recognition difficulties caused by posture changes and lighting conditions. Most of them require the use of special equipment and pictures taken in ideal environments, and Does not take into account the impact caused when the conditions are inconsistent.

The method proposed in the literature A Biometric Identification System Based on Eigenpalm and Eigenfinger Features (biometric identification system based on feature palm and feature finger features) and A multi-matchersystem based on knuckle-based features (multi-matcher system based on knuckle features) It is necessary to use a scanning device to obtain pictures, and the method in the document Online finger-knuckle-print verification for personal authentication (online fingerprint authentication method for personal identification) also requires a special system.

Although the method in the document Palmprint Recognition across Different Devices (palmprint recognition method on different devices) uses mobile devices to collect palm pictures, it requires that the background of the picture must be completely black, and the lighting should be as uniform as possible.

Contents of the invention

The primary purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and to provide a mobile recognition method based on the horizontal stripes of the inner finger. equipment.

Another object of the present invention is to overcome the shortcomings and deficiencies of the prior art, and to provide a mobile recognition device that implements a mobile recognition method based on inner finger stripes. The device mainly includes image acquisition, hand detection, region of interest (ROI) ) positioning, feature extraction and matching of inner finger stripes, etc.

The primary purpose of the present invention is achieved through the following technical solutions: a mobile identification method based on the horizontal stripes of the inner finger, comprising the following steps:

T1: Collect images. In a mobile environment, the user uses the hand image captured by the camera configured on the mobile device.

T2: Hand detection, after processing the skin model based on the mixed Gaussian model and morphological expansion and erosion methods, a complete and accurate hand outline is extracted from the source image.

T3: ROI positioning, using the finger reference point positioning method and Radon projection to further segment four regions of interest from the hand image extracted from T2.

T4: Feature extraction of inner finger stripes, based on competitive coding method to extract feature information of ROI area, including orientation map (orientation map) and energy map (energy map).

T5: feature matching of horizontal stripes of the inner finger. Based on the area histogram statistical method, the two feature maps obtained in the above step T4 are matched with the feature data in the server to obtain the final matching result.

T2: Hand detection mainly includes two steps: rough positioning (T21) and precise positioning (T22):

T21: Rough positioning, using the pre-prepared typical skin color image as the training set, using the improved expectation maximization (EM) algorithm, iteratively calculating the likelihood function through the existing data, so that it converges to a certain optimal value, Thus, the parameters of the mixture Gaussian model are obtained automatically. It is judged whether each pixel of the collected image conforms to a certain Gaussian kernel, if so, it is judged as the hand skin area, otherwise it is judged as the background area, so as to obtain a rough hand image.

T22: Accurate positioning, using morphological methods to process the rough hand image obtained above, can remove small holes and noise inside, so as to obtain a complete and accurate hand image and outline.

T3: After extracting a complete and accurate hand image, four regions of interest are segmented from the image. It is mainly divided into two steps (T31 and T32):

T31: Localization and segmentation of fingers. On the hand image obtained above, we use the distance relationship between the points on the contour line to locate the reference points: five fingertip points and four finger valley points. Based on these reference points, the median axis of each finger is estimated and the area of the finger is extracted. Since the thumb and inner finger stripes contain less information, we only locate the remaining four fingers.

T32: Rotate the finger area image segmented by T31 to the horizontal axis, and crop all samples to a uniform size. Radon projection is performed on the sample, and two obvious peak areas are obtained, which are the areas where the horizontal stripes of the inner finger are located. Then low-pass filtering is used to obtain specific position coordinates to obtain the final ROI area.

T4: Using a method based on competitive coding, features for recognition are extracted from the ROI images obtained in T3. The method is specifically described as follows: We use Gabor filtering to capture the direction information of the image from the image. After obtaining Gabor filters in different directions, the Gabor kernel in each direction is used to perform convolution operation on the source image of the region of interest, and finally n response value images are obtained, corresponding to n different directions, and the direction containing the minimum response value will be selected (dominant) direction, and form an orientation map (orientation map) representing the main direction of each pixel; at the same time, this minimum response value will be stored in another map, after the value is reversed , The quantization operation generates an energy map (energy map) describing the main direction weight of each pixel. Finally, the orientation map and energy map jointly represent the structural feature information of the image.

T5: A feature matching method based on regional histogram statistics, which mainly includes two steps (T51 and T52):

T51: We perform local histogram statistics on the orientation map (orientation map) and energy map (energy map) obtained from T4. We divide the source image into many overlapping regions called patches. Each block consists of several non-overlapping cells. Each cell corresponds to a histogram formed by weighted voting of the internal pixels. The voting results of each grid, that is, the corresponding histogram forms the features of each block, and the features of each block form the feature vector of the entire image, which is used for final recognition.

T52: Calculate the Euclidean distance between the feature vector to be matched and the feature vector stored in the background database. The present invention obtains a threshold value through a machine learning method, and if the Euclidean distance is smaller than the threshold value, the feature matching is successful, and user identity authentication is completed. Due to the method of histogram statistics within the block, the matching algorithm of the present invention is not very sensitive to the positional relationship between the blocks to be matched in the two images within a certain range, so when there is a certain range of differences between the samples to be matched When the translation is dislocated, the similarity between the two pictures can still be effectively expressed, so the present invention has a certain tolerance for dislocation.

Another object of the present invention is achieved through the following technical solutions: a mobile identification device for implementing a mobile identification method based on inner finger stripes, which mainly includes the following modules:

Acquisition module, is used for photographing user's hand image;

A transmission module, configured to transmit the image of the user's hand captured by the acquisition unit to the server through the network;

A preprocessing module, configured to detect the hand region from the user's hand image at the server end, and locate the inner finger stripe region;

The processing module is used to extract and match the user's internal finger stripe information based on the result of the preprocessing unit locating the finger stripe area;

The decision-making module is used to determine the result of verification or identification based on the result of matching the horizontal stripe information of the user's inner finger by the processing unit.

The device used by the acquisition module is a camera equipped on a mobile device, the direction of the camera is perpendicular to the plane formed by the five fingers of the hand, and the shake of the camera is within 20 degrees.

The hand and its background have the following characteristics when the acquisition module collects:

There is a difference in the color of the background and the skin of the hand;

Sufficient light without strong shadows;

The five fingers are naturally flattened and separated;

The preprocessing module includes a hand region detection unit and a positioning unit of inner finger stripes, the hand region detection unit uses a skin model based on a mixed Gaussian model to extract the hand region, and the inner finger stripe location unit uses The geometric information of the hand contour extracts the finger area, performs Radon projection on the finger area, and extracts the inner finger stripe area.

The processing module includes a feature extraction unit and a matching unit of the inner finger stripe region,

The feature extraction unit adopts the feature representation method of the competitive coding; the input of the unit is the region of interest, and the output is the direction map and energy map described in claim 3;

The feature matching unit adopts the local statistics-based method; the input of the unit is the direction map and the energy map, and the output is the Euclidean distance between the input image and the feature vector of the image in the matching database.

The decision-making module makes decisions based on the matching results of a whole hand, and the multi-region-based comprehensive decision-making method includes two modes: a verification mode and a recognition mode;

The verification mode is to carry out one-to-one matching, and judge whether the target image and the image in the database are the same hand;

The recognition mode is one-to-many matching, and the most matching hand is found from the hand database. If the decision energy value is lower than the threshold, it is judged that there is no detected picture of the user's hand in the internal finger stripe database. Here, the weighted average method is used to calculate the decision energy value. Since the horizontal stripes of the little finger usually contain less information, the weight is relatively low. The weights of the index finger, middle finger, ring finger, and little finger are set to 0.3, 0.3, and 0.3, respectively. 0.1, so the total decision-making energy value T=0.3*(T2+T3+T4)+0.1*T5, where T2, T3, T4, and T5 respectively represent the energy values of the index finger, middle finger, ring finger, and little finger.

Working principle of the present invention: the present invention uses the camera of the mobile device to collect hand images, uses the skin model based on the mixed Gaussian model to extract the hand area, and locates the area of the horizontal stripes of the inner finger by analyzing the geometric characteristics of the hand contour line. Competitive coding Gabor filtering technology extracts the features of the inner finger stripes and uses the area histogram to match, thus realizing the identification based on the inner finger stripes biometrics.

Compared with the prior art, the present invention has the following advantages and effects:

1. This method effectively reduces the impact of illumination and user posture on the recognition image by using the relatively stable and obvious inner finger stripes and image processing methods, and overcomes the problems in the literature Human and machine recognition of faces: a survey (human and machine) Face recognition in : A survey) has the disadvantage of requiring a huge database containing a large number of shot samples under different lighting and pose conditions.

2. It is mentioned in the literature Illumination ratio image: synthesizing and recognition with varying illuminations (illumination quotient image: synthesis and recognition under variable illumination), compared with voice, face and palm prints, due to its natural features More suitable for mobile biometric systems. In addition to the smooth surface, the horizontal fingerprint also has rich structural information that is difficult to imitate, and because it is in a small area, its projection transformation is small or even negligible. That is to say, under the condition of low resolution, this method can obtain a higher accuracy rate than palmprint recognition by using the horizontal fingerprint.

3. This method can be implemented only with a mobile device with an ordinary camera, which overcomes the literature ABiometric Identification System Based on Eigenpalm and Eigenfinger Features (a biometric identification system based on feature palm and feature finger features), A multi-matcher system based Onknuckle-based features (multi-matcher system based on knuckle features) and the document Onlinefinger-knuckle-print verification for personal authentication (online fingerprint authentication method for personal identification) require additional restrictions for special equipment.

4. Compared with the method in the document Palmprint Recognition across Different Devices (palmprint recognition method on different devices), which requires that the background of the picture must be completely black, and the illumination should be as uniform as possible, this method uses the distinctive inner finger Horizontal stripes and image processing methods effectively reduce the requirements for image background and lighting environment.

5. Compared with the prior art, the advantage of this method is that it can greatly reduce the condition restrictions when collecting images. The traditional method must use specific auxiliary acquisition tools to collect the recognition image, or have strict requirements on the surrounding environment when collecting the image. However, our method takes advantage of the relatively obvious horizontal stripes of the inner finger, and only needs to use the camera configured on the mobile device as a collection tool. In the process of extracting features, various methods are used to minimize the changes caused by exposure, rotation, and movement of the image under different circumstances, so that it has strong robustness to different environmental conditions.

Description of drawings

Figure 1 is a skin model based on a mixture of Gaussian models.

Fig. 2 is a working flow diagram of the field recognition process of the present invention.

Figure 3a is a rough hand contour image obtained after skin model detection respectively.

Figure 3b is a relatively complete and accurate hand contour image obtained after dilation and erosion morphological processing.

Fig. 4a is a schematic diagram of finger reference point positioning.

Fig. 4b is a schematic diagram of positioning the central axis of the finger.

Fig. 4c is a schematic diagram of finger segmentation.

Fig. 5 is a schematic diagram of the positioning of the region of interest (ROI region) based on the RADON projection of the horizontal stripes of the inner finger.

Fig. 6a is a schematic diagram of a method for scanning a direction map using a local block area.

Figure 6b is a schematic diagram of the block structure.

Fig. 6c is a schematic diagram of bicubic interpolation.

Fig. 7 is a device structure diagram of the present invention.

Fig. 8 is a flowchart of the hand image preprocessing module.

Figure 9 is the ROC diagram based on the ideal environment and the mobile environment.

Detailed ways

The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

Example

The whole identification process of the present invention includes two parts: preparatory work and on-site implementation.

The preparatory work includes two parts: establishing a hand image database and a skin model.

Build a hand image database and collect hand images of target users. The acquisition device is an ordinary camera, and its acquisition conditions mainly include the following characteristics: there is a difference in the color of the background and the skin of the hand; the light is relatively sufficient and there is no strong shadow; the five fingers are naturally flat and separated; within 20 degrees.

The skin model is established, the typical skin color area is extracted from the hand image database as a training set, and the parameters of the mixed Gaussian model are automatically obtained by using the improved expectation maximization algorithm. As shown in Figure 1, the distribution of 8 Gaussian kernels representing the skin model is shown.

The on-site implementation process is shown in Figure 2:

#1 Collect user hand images;

#2 upload the image to the server;

#3 hand extraction;

#4 Finger positioning and segmentation;

#5 Locate the region of interest in the horizontal stripes of the inner finger;

#6 Feature extraction of inner finger stripes;

#7 Match the extracted features in the database;

#8 Comprehensive decision making;

#9 The server returns the matching result to the mobile terminal;

The technical details of each step in the flowchart are as follows:

#1 Collect user’s hand image: The user uses the camera configured on the mobile device to take a hand image. The characteristics of the collection conditions are as follows: there is a difference in the color of the background and the skin of the hand; the light is relatively sufficient without strong shadows; the five fingers are naturally flat, Separate; the direction of the camera is right on the plane of the opponent, and the shaking is within 20 degrees; the palm including the fingers basically occupies the entire image screen.

#2 Upload the image to the server: the mobile device connects to the Internet through 3G or WIFI network, and uploads the collected hand image to the background server.

#3 Hand extraction: Rough positioning of the hand area based on the pre-acquired skin model, specifically described as: judging whether each pixel on the collected image conforms to a Gaussian kernel, and if so, it is judged as the hand skin area, otherwise It is judged as the background area, so as to obtain a rough hand image, as shown in Figure 3a. On this basis, morphological methods are used to remove small holes and noise points in the above rough results, so as to obtain a complete and accurate hand image and contour line, as shown in Figure 3b.

#4 Finger positioning and segmentation: Based on the hand area obtained above, we use the distance relationship between the points on the contour line to locate the reference points: five fingertip points and four finger valley points. As shown in Figure 4a, we start from an endpoint P and count the distance to point P point by point along the contour line. Reduce), select the current point as the feature point, so we can get five fingertip points T1-5, B1, B3, B4, B5 are finger valley points. Not only that, we find the point B2 closest to B3 on the contour line between T2 and B1, and similarly find point B6 as a new feature point, so that we basically determine the approximate position of the finger.

Taking the middle finger as an example, as shown in Figure 4b, we divide the two curves B3T3 and B4T3 into three equal parts to obtain four three-point points M1, M2', M3, and M4', and then start from the point M2 in two directions up and down Traverse each point along the contour line, calculate the distance between this point and M1, get the nearest point M2, similarly get M4, connect M1M2, M3M4, take the connecting line D2V2 of these two key points, which is the central axis of the finger . The finger area can be obtained by making a rectangular frame with the central axis as the axis.

Since the thumb and inner finger stripes contain less information, we only locate the remaining four fingers, as shown in Figure 4c.

#5 Locating the area of interest of the horizontal stripes of the inner finger: Rotate the finger area segmented in the previous step to the level, and do a Radon projection for each finger area. Each finger area can get two peaks, and each finger area uses a low The determined cross-print area of the inner finger is obtained through filtering, and Fig. 5 shows the extraction of a horizontal line of the inner finger in a finger area.

#6 Feature extraction of inner finger stripes: the previous extraction method will fail when the color change is relatively smooth. In order to solve this problem, the present invention uses a competitive coding method to extract the inner finger stripe structure information from the ROI region image . In the present invention, the Gabor filter is used to extract the direction information of the fingerprint stripes. In the present invention, image pixels are convolved at n angles through a Gabor kernel, and the convolution kernel is calculated by a Gabor function based on neurophysiology. The function is:

x'=cosθ·x+sinθ·y,

$\mathrm{<span\; class="notranslate">\; \&kappa;</span>}<span\; class="notranslate">\; =</span>\sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; ln</span>}}<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

$\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; =</span>\frac{\sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&pi;</span>}}}{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; ;</span>$

分别是正弦函数的频率及相位。经过卷积后获得18个响应值图像，然后对比每一个计算单元在18个方向上的敏感程度，响应值越小，表示方向性越强，把响应值最小的方向设为这个计算单元的主方向，把所有计算单元的主方向储存起来，形成描述图像上每个象素点主方向的方向图（orientationmap）。与此同时，这个最小的响应值会被储存到另一幅图,通过对这幅图里面的数据进行减去最小值,取倒数,并量化到[0,1.0]等操作,生成代表每个像素点主方向的权值的能量图（energy map）。energy map的权值越大，代表该像素点在这个主方向上体现出来的线条性就越强。最终orientation map和energy map共同描述图像的结构特征信息。Where θ is the direction of the wavelet, generally 18, that is, a representative direction is taken every 10 degrees; σ is the standard deviation of the Gaussian profile; λ and

are the frequency and phase of the sine function, respectively. After convolution, 18 response value images are obtained, and then the sensitivity of each calculation unit in 18 directions is compared. The smaller the response value, the stronger the directionality, and the direction with the smallest response value is set as the main direction of the calculation unit. Orientation stores the main directions of all computing units to form an orientation map (orientation map) describing the main direction of each pixel on the image. At the same time, the minimum response value will be stored in another graph. By subtracting the minimum value from the data in this graph, taking the reciprocal, and quantizing to [0,1.0], etc., generating a representative of each The energy map of the weight of the main direction of the pixel (energy map). The greater the weight of the energy map, the stronger the linearity of the pixel in this main direction. Finally, the orientation map and energy map jointly describe the structural feature information of the image.

#7 Match the extracted features in the database: After obtaining the structural information (orientation map, energy map) of the horizontal stripes of the inner finger, the present invention uses the method of regional histogram statistics to realize the image structure feature matching that tolerates dislocation. Specifically, it can be described as:

First, define a block area, each block area includes 2x2 subunits, and each subunit includes 4x4 pixels, as shown in Figure 6b.

Second, from left to right, from top to bottom, use a block area to scan the orientation map (orientation map) described in right 6. The step size for moving from left to right is 4 pixels, and the step size for moving from top to bottom is 4 pixels, as shown in Figure 6a.

Third, in each block area, perform histogram-based statistics on the corresponding orientation map (orientation map) area. For each subunit, the respective energy values in the six directions are counted. When calculating the contribution of each pixel point to the six different directions of different subunits, the weight must be bicubically interpolated according to the spatial position and direction angle of the pixel point, as shown in Figure 6c. The weight here refers to the value corresponding to the pixel on the energy map (energy map). Thus, each subunit contains a 6-dimensional feature vector, which also means that each block contains a 24-dimensional feature vector. Combining the feature vectors of each block area forms a feature vector describing the entire image.

Fourth, to calculate the structural similarity of the two images is to calculate the distance between the feature vectors of the two images, using the Euclidean distance calculation method. Through a large amount of experimental data, we predict a reliable distance as a threshold, based on which it is judged whether the horizontal stripes of the two inner fingers match. The threshold value mentioned here is an empirical value based on a large number of experimental tests.

#8 Comprehensive decision-making: In order to obtain a more reliable matching structure, the present invention adopts a comprehensive decision-making method based on multiple regions, which can be specifically described as: for each hand, the matching region includes the third finger of the four fingers except the thumb. The interdigital stripes of the nodes, that is, a hand contains four matching subregions. As long as two or more matching results of the four sub-regions are judged as matching, it is considered that the corresponding images of the two hands are from the same hand. The decision energy value is the weighted average of the matching energy values of the interfinger stripe region on the match.

#9 The server returns the matching result to the mobile terminal: the server transmits the matching result to the corresponding mobile terminal through the network, and the matching result is displayed on the mobile terminal.

Based on the above process, the equipment required for the unrestricted movement identification method based on the horizontal stripes of the inner finger proposed in the present invention includes 5 modules, as shown in FIG. 7 . The equipment consists of acquisition module, transmission module, preprocessing module, processing module and decision module.

The acquisition module is responsible for taking images of the user's hands;

The transmission module is responsible for transmitting the image of the user's hand captured by the acquisition unit to the server through the network;

The preprocessing module is responsible for detecting the hand area from the collected images, and locating the ROI area of the horizontal stripes of the inner finger;

The processing module is responsible for feature extraction and matching of the obtained internal finger stripe ROI region;

The decision module is responsible for deciding the result of verification or recognition based on the matching data.

The preprocessing module includes a hand area detection unit and a positioning unit for inner finger stripes, characterized in that: the hand area detection unit uses a skin model based on a mixed Gaussian model to extract the hand area, and the inner finger stripes are positioned The unit uses the geometric information of the hand contour to extract the finger area, performs Radon projection on the finger area, and extracts the ROI area of the horizontal stripes of the inner finger. The specific process is shown in Figure 8.

The processing module includes a feature extraction unit and a matching unit for the inner finger stripe area, wherein the feature extraction unit adopts the feature representation method of competitive encoding, the input is the ROI area image of the inner finger stripe area, and the output is an orientation map (orientation map) and an energy map (energy map); the feature matching unit adopts a method based on local statistics, the input is orientation map and energy map, and the output is the Euclidean distance between the target image and the feature vector of the image in the database.

The decision module includes a verification unit and an identification unit. The decision-making module is based on a multi-regional comprehensive decision-making model. Each hand is matched with 4 ROI regions, which are the third knuckle of the middle index finger, middle finger, ring finger, and little finger. The verification unit is a one-to-one matching, judging whether the target image and the image in the database are the same hand; the recognition unit is a one-to-many matching, finding the best matching hand from the hand database, that is, the hand with the highest decision energy value, If the decision energy value is below a threshold, it is considered that no picture of the same hand exists in the database.

In order to verify the method of the present invention, we obtained samples for matching experiments in an ideal environment with consistent illumination and in a mobile environment, as shown in Table 1 and Figure 9 are the experimental results of the present invention:

sample type Number of samples correct match wrong match Match success rate ideal environment 1000 992 8 99.2% mobile environment 500 485 15 97.0%

Table 1

In the mobile environment, 22->15,

It can be seen from the experimental results that the recognition of the present invention in a mobile environment can still maintain a good matching effect, so compared with the previous method, our invention is more suitable for practical application in mobile devices.

What Fig. 9 represented is the receiver operating characteristic curve of the present invention under ideal environment and mobile environment, and wherein GAR represents correct acceptance rate, and FAR represents false acceptance rate, as can be seen from the figure, under ideal environment our The method can obtain a high GAR value with a very low FAR, and the comprehensive matching rate exceeds 99%. Even in a mobile environment, we can easily achieve a matching success rate of 97%, which shows that the performance of the present invention meets the usage requirements.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (10)

1. A mobile identification method based on inner finger stripes, comprising the following steps: Step 1. Establishing a user internal finger stripe database; Step 2. Automatically detect the hand area; Step 3. Locate the horizontal stripe area of the inner finger; Step 4. Feature representation robust to illumination, pose and viewpoint; Step 5. Feature matching that is tolerant to dislocation; Step 6. Based on multi-region comprehensive decision-making. 2. the mobile identification method based on inner finger stripes according to claim 1, characterized in that, In the step 1, the database of horizontal fingerprints of the user is established by using a camera on the mobile device to capture images of the user's hand when the finger moves; In the step 3, the method for locating the horizontal stripes region of the inner finger comprises the following steps: The first step: set up a mixed Gaussian model generated by the skin color image as a training set, improve the expectation maximization algorithm to obtain the parameters of each Gaussian distribution in the mixed Gaussian model, and build a skin model; Step 2: Substituting each pixel color value of the captured hand image into each Gaussian distribution of the established skin model, if the pixel value conforms to the Gaussian distribution, it is judged as a hand region, otherwise it is a background region; Step 3: filter the captured image by means of dilation and erosion in morphology, eliminate fragmented areas, and obtain a complete hand outline; Step 4: use the distance relationship between the points on the complete hand contour line to locate the reference point; Step 5: use the geometric relationship between the reference point and the contour line to determine the position and central axis of each finger in the captured image; Step 6: Analyze the captured image with Radon projection to obtain the region of interest. 3. The movement recognition method based on inner finger stripes according to claim 1, characterized in that, in said step 4, said feature representation with robustness to illumination, posture and viewpoint refers to a method based on competitive coding A feature representation method, the feature representation method based on competitive coding comprises the following steps: A. For each pixel point in the inner finger stripe area, calculate the energy value of the Gabor filter in n directions, and the direction with the smallest energy value is set as the main direction of the pixel point; the n directions refer to 360 degrees Divided into n intervals, and n is an integer greater than 18; B. An image of horizontal stripes of the inner finger is competitively encoded to output two feature images of the same resolution; the image representing the main direction of each pixel is called a direction map; the weight representing the main direction of each pixel is called Make an energy diagram. 4. The mobile recognition method based on the inner finger stripes according to claim 1, characterized in that, in the step 5, the dislocation-tolerant feature matching method refers to a feature matching method based on local statistics, The feature matching method based on local statistics comprises the following steps: (1) Define a block area, which contains the square subunits of c, and each subunit contains the square pixels of S; (2) On the direction diagram, scan from left to right, from top to bottom, with a block area; the step size of moving from left to right is d1, and the step size of moving from top to bottom is d2, the step size d1 and d2 must ensure that at least 1/3 of the area overlaps between the block areas; each scan, perform histogram-based statistics on the space corresponding to the block area; perform statistics on the histogram of the block area; In the step (2), the statistical method for performing statistics on the histogram of the block area is: ① For each subunit, count the energy values in m directions respectively, m must be less than n described in claim 3; ② When calculating the contribution of each pixel point to m different directions of different subunits, perform bicubic interpolation on the weight value according to the spatial position and direction angle of the pixel point, and the weight value is corresponding to the pixel on the energy map value; ③ The feature vector dimension of each subunit is n, and there are c units in total, then the feature vector dimension of each block area is the product of the square of c and n; ④ The feature vectors of each block area are combined to form a feature vector describing the entire image; ⑤Use the Euclidean distance calculation method to calculate the distance between the target image and the image feature vector in the database, and calculate the structural similarity between the images to be matched. 5. the mobile identification method based on internal finger stripes according to claim 1, is characterized in that, in described step 6, the method for the comprehensive decision-making based on multi-region comprises the following steps: 1. Set the index finger, middle finger, ring finger, and little finger of the four fingers of the detected hand as the horizontal stripes of the inner finger as the matching sub-region; II. Perform the feature representation based on competitive encoding and feature matching based on local statistics for each sub-region; Ⅲ. Compare the sub-regions of the index finger, middle finger, ring finger and little finger with the data in the user's inner finger stripe database, and if there are two or more fingers that match the data in the user's inner finger stripe database , then it is determined that the image of the detected hand matches the image of the matched hand in the database; Ⅳ. The decision energy value is the weighted average of the matching energy values of the inner finger stripe area where the index finger, middle finger, ring finger and little finger match the data in the user's inner finger stripe database. 6. A mobile identification device for realizing a mobile identification method based on inner finger stripes, characterized in that it comprises: Acquisition module, is used for photographing user's hand image; A transmission module, configured to transmit the image of the user's hand captured by the acquisition unit to the server through the network; A preprocessing module, configured to detect the hand region from the user's hand image at the server end, and locate the inner finger stripe region; The processing module is used to extract and match the user's internal finger stripe information based on the result of the preprocessing unit locating the finger stripe area; The decision-making module is used to determine the result of verification or identification based on the result of matching the horizontal stripe information of the user's inner finger by the processing unit. 7. The mobile identification device realizing the mobile identification method based on the horizontal stripes of the inner finger according to claim 6, characterized in that: the device used by the acquisition module is a camera equipped on a mobile device, and the direction of the camera is in line with that of the hand The plane formed by the spread of five fingers is vertical. 8. the mobile identification device that realizes the mobile identification method based on the horizontal stripes of the inner finger according to claim 6, wherein the preprocessing module includes a hand area detection unit and a positioning unit of the horizontal stripes of the inner finger, the The hand area detection unit uses a skin model based on a mixed Gaussian model to extract the hand area, and the inner finger stripe location unit uses the geometric information of the hand contour to extract the finger area, and performs Radon projection on the finger area to extract the inner finger stripes area. 9. The mobile identification device realizing the mobile identification method based on the horizontal stripes of the inner finger according to claim 6, wherein the processing module includes a feature extraction unit and a matching unit of the horizontal stripes of the inner finger region, The feature extraction unit adopts the feature representation method of the competitive coding; the input of the unit is the described region of interest, and the output is the direction diagram and energy diagram described in claim 3; The feature matching unit adopts the method based on local statistics; the input of the unit is the direction map and the energy map, and the output is the Euclidean distance between the input image and the feature vector of the image in the database for matching. 10. The mobile identification device for realizing the mobile identification method based on the inner finger stripes according to claim 6, characterized in that: the decision-making module makes decisions based on the matching results of the hands, and the multi-region-based comprehensive Decision-making method, including two modes: verification mode and recognition mode; The verification mode is to carry out one-to-one matching, that is: judge whether the image of the detected hand and the image of the hand in the internal finger stripe database are the same hand; The recognition mode is a one-to-many matching, that is: find the image of the most matching hand from the database of inner finger stripes, if the decision energy value is lower than the threshold, it is determined that the detected data does not exist in the inner finger stripe database. A picture of the user's hand.

Images